---

---

CASE STUDY - 1 :: Healthcare Provider Fraudulent Detection

---

---

This notebook contains the extensive data analysis that been carried out on the publicly available dataset at Kaggle with the intent:

• To gain better understanding of the dataset from different levels.
• To explore the dataset in order to uncover the potential hidden patterns.
• And, to find the answers of various WHYs.

Kindly checkout below link for gaining BUSINESS related insights about this problem:

• Deck : Detailed Explanation

Kindly checkout below link for TECHNICAL description about this problem:

• Technical Document

Notebook Contents

CASE STUDY - 1 :: Healthcare Provider Fraudulent Detection

Notebook Contents

Downloading_Train_Data_Files

Importing_Libraries

Importing_Dataset

Exploring_BENEFICIARY_Data

Q1. How many unique beneficiaries we have in our dataset?

Q2. How many records we have at the GENDER level?

Q3. Lets calculate the AGE of every BENEFICIARY?

Q4. Lets see the ratio of GENDER across various HUMAN RACE?

Q5. Lets see the number of beneficiaries with Chronic Renal Disease.

Q6. Lets see the number of beneficiaries on the basis of State Codes.

Q7. Lets see the number of beneficiaries on the basis of Country Codes.

Q8.1 Lets see the number of beneficiaries on the basis of 'NoOfMonths_PartACov'.

Q8.2 Lets see the number of beneficiaries on the basis of 'NoOfMonths_PartBCov'.

Q9 Lets see the number of beneficiaries on the basis of 'ChronicCond_Alzheimer'. And, the Annual IP & OP expenditures for such patients.

Q10. Lets see the number of beneficiaries on the basis of 'ChronicCond_Heartfailure'. And, the Annual IP & OP expenditures for such patients.

Q11. Lets see the number of beneficiaries on the basis of 'ChronicCond_KidneyDisease'. And, the Annual IP & OP expenditures for such patients.

Q12. Lets see the number of beneficiaries on the basis of 'ChronicCond_Cancer'. And, the Annual IP & OP expenditures for such patients.

Q13. Lets see the number of beneficiaries on the basis of 'ChronicCond_ObstrPulmonary'. And, the Annual IP & OP expenditures for such patients.

Q14. Lets see the number of beneficiaries on the basis of 'ChronicCond_Depression'. And, the Annual IP & OP expenditures for such patients.

Q15. Lets see the number of beneficiaries on the basis of 'ChronicCond_Diabetes'. And, the Annual IP & OP expenditures for such patients.

Q16. Lets see the number of beneficiaries on the basis of 'ChronicCond_IschemicHeart'. And, the Annual IP & OP expenditures for such patients.

Q17. Lets see the number of beneficiaries on the basis of 'ChronicCond_Osteoporasis'. And, the Annual IP & OP expenditures for such patients.

Q18. Lets see the number of beneficiaries on the basis of 'ChronicCond_rheumatoidarthritis'. And, the Annual IP & OP expenditures for such patients.

Q19. Lets see the number of beneficiaries on the basis of 'ChronicCond_stroke'. And, the Annual IP & OP expenditures for such patients.

Q20. Lets see the number of beneficiaries on the basis of 'RenalDiseaseIndicator'. And, the Annual IP & OP expenditures for such patients.

Q21. Lets check the percentiles of the pre-disease indicators for the Annual IP expenditures for such patients.

Q22. Lets just visualize the spread of pre-disease indicators for the Annual IP and OP expenditures across males and females.

Q23. Lets visualize the spread of Annual IP and OP expenditures through out the AGE and its assciated features for males and females.

SUMMARY

Downloading_Train_Data_Files

## TRAIN set files
!gdown 12zSQN2FOxmuXFhz2xzPNussPisEfVP5w
!gdown 13XyBakfHiG-BNQPrYFXAHlsOcfICOTpx
!gdown 1dLxl4vkykPcm4Zj0abYR0Ohr7STQHg-1
!gdown 1rFER-7VuYb7GfCYeJrfxPidgK0lwqw3R

Importing_Libraries

import os
import sys
import math
import scipy as scipy
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

%matplotlib inline

pd.set_option('display.max_columns',30)
label_font_dict = {'family':'sans-serif','size':13.5,'color':'brown','style':'italic'}
title_font_dict = {'family':'sans-serif','size':16.5,'color':'Blue','style':'italic'}

Importing_Dataset

train_bene_df = pd.read_csv("Dataset\Train\Train_Beneficiarydata-1542865627584.csv")
train_ip_df = pd.read_csv("Dataset\Train\Train_Inpatientdata-1542865627584.csv")
train_op_df = pd.read_csv("Dataset\Train\Train_Outpatientdata-1542865627584.csv")

Exploring_BENEFICIARY_Data

train_bene_df.shape

(138556, 25)

train_bene_df.columns

Index(['BeneID', 'DOB', 'DOD', 'Gender', 'Race', 'RenalDiseaseIndicator',
       'State', 'County', 'NoOfMonths_PartACov', 'NoOfMonths_PartBCov',
       'ChronicCond_Alzheimer', 'ChronicCond_Heartfailure',
       'ChronicCond_KidneyDisease', 'ChronicCond_Cancer',
       'ChronicCond_ObstrPulmonary', 'ChronicCond_Depression',
       'ChronicCond_Diabetes', 'ChronicCond_IschemicHeart',
       'ChronicCond_Osteoporasis', 'ChronicCond_rheumatoidarthritis',
       'ChronicCond_stroke', 'IPAnnualReimbursementAmt',
       'IPAnnualDeductibleAmt', 'OPAnnualReimbursementAmt',
       'OPAnnualDeductibleAmt'],
      dtype='object')

train_bene_df.dtypes

BeneID                             object
DOB                                object
DOD                                object
Gender                              int64
Race                                int64
RenalDiseaseIndicator              object
State                               int64
County                              int64
NoOfMonths_PartACov                 int64
NoOfMonths_PartBCov                 int64
ChronicCond_Alzheimer               int64
ChronicCond_Heartfailure            int64
ChronicCond_KidneyDisease           int64
ChronicCond_Cancer                  int64
ChronicCond_ObstrPulmonary          int64
ChronicCond_Depression              int64
ChronicCond_Diabetes                int64
ChronicCond_IschemicHeart           int64
ChronicCond_Osteoporasis            int64
ChronicCond_rheumatoidarthritis     int64
ChronicCond_stroke                  int64
IPAnnualReimbursementAmt            int64
IPAnnualDeductibleAmt               int64
OPAnnualReimbursementAmt            int64
OPAnnualDeductibleAmt               int64
dtype: object

train_bene_df.head()

	BeneID	DOB	DOD	Gender	Race	RenalDiseaseIndicator	State	County	NoOfMonths_PartACov	NoOfMonths_PartBCov	ChronicCond_Alzheimer	ChronicCond_Heartfailure	ChronicCond_KidneyDisease	ChronicCond_Cancer	ChronicCond_ObstrPulmonary	ChronicCond_Depression	ChronicCond_Diabetes	ChronicCond_IschemicHeart	ChronicCond_Osteoporasis	ChronicCond_rheumatoidarthritis	ChronicCond_stroke	IPAnnualReimbursementAmt	IPAnnualDeductibleAmt	OPAnnualReimbursementAmt	OPAnnualDeductibleAmt
0	BENE11001	1943-01-01	NaN	1	1	0	39	230	12	12	1	2	1	2	2	1	1	1	2	1	1	36000	3204	60	70
1	BENE11002	1936-09-01	NaN	2	1	0	39	280	12	12	2	2	2	2	2	2	2	2	2	2	2	0	0	30	50
2	BENE11003	1936-08-01	NaN	1	1	0	52	590	12	12	1	2	2	2	2	2	2	1	2	2	2	0	0	90	40
3	BENE11004	1922-07-01	NaN	1	1	0	39	270	12	12	1	1	2	2	2	2	1	1	1	1	2	0	0	1810	760
4	BENE11005	1935-09-01	NaN	1	1	0	24	680	12	12	2	2	2	2	1	2	1	2	2	2	2	0	0	1790	1200

Q1. How many unique beneficiaries we have in our dataset?

train_bene_df['BeneID'].nunique()

138556

Q2. How many records we have at the GENDER level?

train_bene_df['Gender'].unique()

array([1, 2], dtype=int64)

train_bene_df['Gender'] = train_bene_df['Gender'].apply(lambda val: 0 if val == 2 else 1)

# Here, I'm displaying the distribution of BENEFICIARIES on the basis of GENDER 
with plt.style.context('seaborn'):
  plt.figure(figsize=(10,8))
  fig = train_bene_df['Gender'].value_counts().plot(kind='bar', color=['yellow','purple'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.015), ha='center', fontsize=13.5)
  # Providing the labels and title to the graph
  plt.xlabel("Gender Code", fontdict=label_font_dict)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Distribution of BENEFICIARIES based on GENDER\n", fontdict=title_font_dict)

OBERVATION

• From the above plot, we can decude that the ratio b/w GENDER_0 : GENDER_1 is 57% : 43%.

Q3. Lets calculate the AGE of every BENEFICIARY?

train_bene_df['DOB'] = pd.to_datetime(train_bene_df['DOB'], format="%Y-%m-%d")

train_bene_df['Patient_Age_Year'] = train_bene_df['DOB'].dt.year
train_bene_df['Patient_Age_Month'] = train_bene_df['DOB'].dt.month

• Adding new feature :::::::::::::::::::::::: "YEAR of birth of beneficiaries"

bene_age_year_df = pd.DataFrame(train_bene_df['Patient_Age_Year'].value_counts()).reset_index(drop=False)
bene_age_year_df.columns= ['year','num_of_beneficiaries']
bene_age_year_df = bene_age_year_df.sort_values(by='year')

# Here, I'm displaying the distribution of BENEFICIARIES on the basis of their YEAR of Birth 
with plt.style.context('seaborn'):
  plt.figure(figsize=(21,9))
  fig = sns.barplot(data=bene_age_year_df, x='year', y='num_of_beneficiaries', palette='inferno')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],1))+"%"}', (x + width/2, y + height*1.025), ha='center', fontsize=13.5, rotation=90)
  # Providing the labels and title to the graph
  plt.xlabel("\nBeneficary YEAR of Birth", fontdict=label_font_dict)
  plt.xticks(rotation=90)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.minorticks_on()
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.title("Distribution of BENEFICIARIES based on their YEAR of birth\n", fontdict=title_font_dict)

OBERVATION

• From the above plot, we can decude that the majority of the beneficiaries are from the YEAR - 1919 to 1943.
  • More specifically, overall the percentage of number of beneficiaries with YEAR of birth from 1939 to 1943 is the highest. Whereas, it is lowest from 1978 to 1983.

REASONING

• The reason for adding the "YEAR of Birth of Beneficiaries" as a new feature in the dataset with the intent to find whether personal details has been forged by the beneficiary or its direct/indirect association with Fraud claims.

• Adding new feature :::::::::::::::::::::::: "MONTH of birth of beneficiaries"

bene_age_month_df = pd.DataFrame(train_bene_df['Patient_Age_Month'].value_counts()).reset_index(drop=False)
bene_age_month_df.columns= ['month','num_of_beneficiaries']
bene_age_month_df = bene_age_month_df.sort_values(by='month')

# Here, I'm displaying the distribution of BENEFICIARIES on the basis of their MONTH of Birth 
with plt.style.context('seaborn'):
  plt.figure(figsize=(12,8))
  fig = sns.barplot(data=bene_age_month_df, x='month', y='num_of_beneficiaries', palette='summer')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],1))+"%"}', (x + width/2, y + height*1.025), ha='center', fontsize=13.5, rotation=90)
  # Providing the labels and title to the graph
  plt.xlabel("\nBeneficary MONTH of Birth\n", fontdict=label_font_dict)
  plt.xticks(rotation=90)
  plt.ylabel("Number or % share of patients", fontdict=label_font_dict)
  plt.minorticks_on()
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.title("Distribution of BENEFICIARIES based on their MONTH of birth\n", fontdict=title_font_dict)

OBERVATION

• From the above plot, we can decude that there is no such difference b/w the patients based on their MONTH of birth.
  • Initial, look suggests me that this feature might not be more of a use.

REASONING

• The reason for adding the "MONTH of Birth of Beneficiaries" as a new feature in the dataset with the intent to find whether personal details has been forged by the beneficiary or its direct/indirect association with Fraud claims.

train_bene_df.head()

	BeneID	DOB	DOD	Gender	Race	RenalDiseaseIndicator	State	County	NoOfMonths_PartACov	NoOfMonths_PartBCov	ChronicCond_Alzheimer	ChronicCond_Heartfailure	ChronicCond_KidneyDisease	ChronicCond_Cancer	ChronicCond_ObstrPulmonary	ChronicCond_Depression	ChronicCond_Diabetes	ChronicCond_IschemicHeart	ChronicCond_Osteoporasis	ChronicCond_rheumatoidarthritis	ChronicCond_stroke	IPAnnualReimbursementAmt	IPAnnualDeductibleAmt	OPAnnualReimbursementAmt	OPAnnualDeductibleAmt	Patient_Age_Year	Patient_Age_Month
0	BENE11001	1943-01-01	NaN	1	1	0	39	230	12	12	1	2	1	2	2	1	1	1	2	1	1	36000	3204	60	70	1943	1
1	BENE11002	1936-09-01	NaN	0	1	0	39	280	12	12	2	2	2	2	2	2	2	2	2	2	2	0	0	30	50	1936	9
2	BENE11003	1936-08-01	NaN	1	1	0	52	590	12	12	1	2	2	2	2	2	2	1	2	2	2	0	0	90	40	1936	8
3	BENE11004	1922-07-01	NaN	1	1	0	39	270	12	12	1	1	2	2	2	2	1	1	1	1	2	0	0	1810	760	1922	7
4	BENE11005	1935-09-01	NaN	1	1	0	24	680	12	12	2	2	2	2	1	2	1	2	2	2	2	0	0	1790	1200	1935	9

• Adding new indicator :::::::::::::::::::::::: "Beneficiary Dead or Alive?"
  • This represent whether a patient is alive or not?
    • Here, the point to understand is that there can be ONE to MANY relationship b/w a BENEFICIARY and CLAIM filed, thus I'm making an assumption here is that beneficiaries with DOD as NA are alive, however this cannot be always true once we join it with CLAIMS data.
      • For example, A patient gets successfully operated for Kidney Failure on Sep 2012 and filed CLAIM-I. So, this claimed gets approved with DOD as NA.
        • Same patient gets operated after 4 months for Cardiac Failure but couldn't survived and filed CLAIM-II. Thus, for CLAIM-I this person shoud have DEAD_or_ALIVE indicator as FALSE whereas for CLAIM-II it should be TRUE.

• REASONING
  • My intention behind adding this feature is to see whether FRAUD claims are higher for DEAD patients or not?
    • The reason here is that generally when an organized fraud happens then healthcare providers falsely diagnose, misdiagnose, or overdiagnose a disease, which can lead to harmful courses of medication or treatments or procedures being prescribed or performed. All of this together can lead to patient death.

# 0 means ALIVE and 1 means DEAD
train_bene_df['Dead_or_Alive'] = train_bene_df['DOD'].apply(lambda val: 0 if val != val else 1)

train_bene_df['Dead_or_Alive'].value_counts()

0    137135
1      1421
Name: Dead_or_Alive, dtype: int64

# Here, I'm displaying the distribution of whether BENEFICIARY is ALIVE or NOT?
with plt.style.context('seaborn'):
  plt.figure(figsize=(10,8))
  fig = train_bene_df['Dead_or_Alive'].value_counts().plot(kind='bar', color=['lightgreen','coral'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.015), ha='center', fontsize=13.5)
  # Providing the labels and title to the graph
  plt.xlabel("Alive or Dead Status?", fontdict=label_font_dict)
  plt.xticks(labels=["ALIVE","DEAD"], ticks=[0,1], rotation=20)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Distribution of BENEFICIARIES based on Alive or Dead Status\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph tells us that almost 99% of the beneficiaries are ALIVE and very small percentage of patients are DEAD. It would be good to see how much this feature has impact on the actual label.

train_bene_df['DOD'] = pd.to_datetime(train_bene_df['DOD'])

# Greatest Date of Death in the TRAIN set for beneficiaries
max_bene_DOD = max(train_bene_df['DOD'].unique()[1:])
max_bene_DOD

numpy.datetime64('2009-12-01T00:00:00.000000000')

# For all NAN DODs filling the greatest Date of Death
train_bene_df['DOD'].fillna(value=max_bene_DOD, inplace=True)

• Adding new feature :::::::::::::::::::::::: "Beneficiary AGE"
  • This represent the AGE of the beneficiary.

• REASONING
  • My intention behind adding this feature is to better analyse the CLAIM and Patient related data.

train_bene_df['AGE'] = np.round(((train_bene_df['DOD'] - train_bene_df['DOB']).dt.days)/365.0,1)

train_bene_df.drop(labels=['DOD'],axis=1,inplace=True)

train_bene_df.head()

	BeneID	DOB	Gender	Race	RenalDiseaseIndicator	State	County	NoOfMonths_PartACov	NoOfMonths_PartBCov	ChronicCond_Alzheimer	ChronicCond_Heartfailure	ChronicCond_KidneyDisease	ChronicCond_Cancer	ChronicCond_ObstrPulmonary	ChronicCond_Depression	ChronicCond_Diabetes	ChronicCond_IschemicHeart	ChronicCond_Osteoporasis	ChronicCond_rheumatoidarthritis	ChronicCond_stroke	IPAnnualReimbursementAmt	IPAnnualDeductibleAmt	OPAnnualReimbursementAmt	OPAnnualDeductibleAmt	Patient_Age_Year	Patient_Age_Month	Dead_or_Alive	AGE
0	BENE11001	1943-01-01	1	1	0	39	230	12	12	1	2	1	2	2	1	1	1	2	1	1	36000	3204	60	70	1943	1	0	67.0
1	BENE11002	1936-09-01	0	1	0	39	280	12	12	2	2	2	2	2	2	2	2	2	2	2	0	0	30	50	1936	9	0	73.3
2	BENE11003	1936-08-01	1	1	0	52	590	12	12	1	2	2	2	2	2	2	1	2	2	2	0	0	90	40	1936	8	0	73.4
3	BENE11004	1922-07-01	1	1	0	39	270	12	12	1	1	2	2	2	2	1	1	1	1	2	0	0	1810	760	1922	7	0	87.5
4	BENE11005	1935-09-01	1	1	0	24	680	12	12	2	2	2	2	1	2	1	2	2	2	2	0	0	1790	1200	1935	9	0	74.3

# Here, I'm displaying the distribution of AGE of Beneficiaries?
with plt.style.context('seaborn'):
  plt.figure(figsize=(10,8))
  train_bene_df['AGE'].plot(kind='hist', color='purple')
  # Providing the labels and title to the graph
  plt.xlabel("\nBeneficiaries Age in years", fontdict=label_font_dict)
  plt.ylabel("Frequency of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Distribution of BENEFICIARIES AGE", fontdict=title_font_dict)
  plt.legend();

• OBSERVATION
  • The above graph tells us that the majority of the beneficiaries are b/w 65 to 85 years.

train_bene_df['AGE'].describe()

count    138556.000000
mean         73.632756
std          12.736605
min          26.000000
25%          68.000000
50%          74.300000
75%          82.000000
max         101.000000
Name: AGE, dtype: float64

• Adding new categorical column :::::::::::::::::::::::: "Beneficiary AGE brackets"
  • This will represent whether a patient falls in a specific AGE bracket?
    • [1-40] --> Group 1 --> Young
    • (40-60] --> Group 2 --> Mid Aged
    • (60-80] --> Group 3 --> Old
    • (80 or more] --> Group 4 --> Very Old

• REASONING
  • My intention behind adding this feature is to see whether FRAUD claims are higher for specific AGE Groups?
    • The reasoning here is that there may be a potential pattern that providers files higher number of FRAUD claims for either a younger age group or very old age group.

def bene_age_brackets(val):
  if val >=1 and val <=40:
    return 'Young'
  elif val > 40 and val <=60:
    return 'Mid'
  elif val > 60 and val <= 80:
    return 'Old'
  else:
    return 'Very Old'

train_bene_df['AGE_groups'] = train_bene_df['AGE'].apply(lambda age: bene_age_brackets(age))

age_grps = list(train_bene_df['AGE_groups'].unique())
for grp in age_grps:
  # Here, I'm displaying the distribution of AGE GROUPS of Beneficiaries?
  with plt.style.context('seaborn'):
    plt.figure(figsize=(8,6))
    train_bene_df[train_bene_df['AGE_groups'] == grp]['AGE'].plot(kind='hist', color='grey')
    # Providing the labels and title to the graph
    plt.xlabel("\nBeneficiaries Age in years", fontdict=label_font_dict)
    plt.ylabel("Frequency of patients\n", fontdict=label_font_dict)
    plt.grid(which='major', linestyle="--", color='lightgrey')
    plt.minorticks_on()
    plt.title("Distribution of BENEFICIARIES Age group -- {}".format(str(grp).upper()), fontdict=title_font_dict)
    plt.legend();

• OBSERVATION
  • The above graphs tells us about the spread of beneficiaries across the Age Groups.
    • For YOUNG group we can say that the spread is quite even across the ages of beneficiaries.

Q4. Lets see the ratio of GENDER across various HUMAN RACE?

• The world population can be divided into 4 major races:
  • white/Caucasian
  • Mongoloid/Asian
  • Negroid/Black
  • Australoid.

This is based on a racial classification made by Carleton S. Coon in 1962. Refer here.

train_bene_df['Race'].unique()

array([1, 2, 3, 5], dtype=int64)

# Here, I'm displaying the distribution of BENEFICIARIES on the basis of Human RACE
with plt.style.context('seaborn'):
  plt.figure(figsize=(10,8))
  fig = train_bene_df['Race'].value_counts().plot(kind='bar', color=['lightgreen','coral','purple','red'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.02), ha='center', fontsize=13.5)
  # Providing the labels and title to the graph
  plt.xlabel("Human RACE", fontdict=label_font_dict)
  plt.xticks(labels=["Race_1","Race_2","Race_3","Race_5"], ticks=[0,1,2,3], rotation=10)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Distribution of BENEFICIARIES on the basis of Human RACE\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph tells us that there is a serious imbalance in the records for Human Race categories.

# Lets validate whether we have imbalance of males and females across the human races
with plt.style.context('seaborn'):
  plt.figure(figsize=(12,8))
  fig = train_bene_df.groupby(['Gender','Race'])['AGE'].count().plot(kind='bar', color=['lightgreen','coral','purple','red','lightgreen','coral','purple','red'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.02), ha='center', fontsize=13.5)
  # Providing the labels and title to the graph
  plt.xlabel("Gender & Human RACE", fontdict=label_font_dict)
  plt.xticks(labels=[("Gender_0","Race_1"), ("Gender_0","Race_2"), ("Gender_0","Race_3"), ("Gender_0","Race_5"),("Gender_1","Race_1"), ("Gender_1","Race_2"), ("Gender_1","Race_3"), ("Gender_1","Race_5")],
             ticks=[0,1,2,3,4,5,6,7], rotation=80)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Imbalance of males and females across the human races?\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us whether we have an imbalance of males or females across the human races.
    • And, it looks like within a specific human race there is no such gender imbalance.

Q5. Lets see the number of beneficiaries with Chronic Renal Disease.

• There are two main types of kidney disease - short-term (acute kidney injury) and lifelong (chronic).

• The two main types of kidney disease are short-term (acute kidney injury) and lifelong (chronic kidney disease).

  • Chronic kidney disease, also known as chronic renal disease or CKD, is a condition characterized by a gradual loss of kidney function over time.

• What are the main causes of chronic kidney disease?

  • Diabetes and high blood pressure, or hypertension, are responsible for two-thirds of chronic kidney disease cases.

For more info refer here --> link-1 link-2

# Here, I'm displaying the distribution of BENEFICIARIES on the basis of State
with plt.style.context('seaborn'):
  plt.figure(figsize=(10,8))
  fig = train_bene_df['RenalDiseaseIndicator'].value_counts().plot(kind='bar', color=['green','orange'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.02), ha='center', fontsize=13.5, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nRenal Disease present or not?", fontdict=label_font_dict)
  plt.xticks(labels=["NO","YES"], ticks=[0,1], rotation=10, size=12)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Distribution of BENEFICIARIES on the basis of Renal Disease\n", fontdict=title_font_dict)

  # 0 means NO and 1 means YES
  print(pd.DataFrame(train_bene_df['RenalDiseaseIndicator'].value_counts()),"\n")

   RenalDiseaseIndicator
0                 118978
Y                  19578 

• OBSERVATION
  • Above graph tells us that around 14% of beneficiaries has or had Kidney Failure(Renal Disease).

# Here, I'm displaying the distribution of BENEFICIARIES on the basis of "ChronicCond_KidneyDisease"
with plt.style.context('seaborn'):
  plt.figure(figsize=(10,8))
  fig = train_bene_df['ChronicCond_KidneyDisease'].value_counts().plot(kind='bar', color=['green','orange'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.02), ha='center', fontsize=13.5, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\n Chronic Kidney Disease present or not?", fontdict=label_font_dict)
  plt.xticks(labels=["NO","YES"], ticks=[0,1], rotation=10, size=12)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Distribution of BENEFICIARIES on the basis of 'ChronicCond_KidneyDisease'\n", fontdict=title_font_dict)

# I believe 2 means NO and 1 means YES
print(pd.DataFrame(train_bene_df['ChronicCond_KidneyDisease'].value_counts()),'\n')

   ChronicCond_KidneyDisease
2                      95277
1                      43279 

• OBSERVATION
  • Above graph tells us that around 31% of Beneficiaries had Chronic Kidney Disease.

• In the beneficiary dataset, we have 2 columns:
  • RenalDisease
  • ChronicCond_KidneyDisease Both of these columns appears to be representing the beneficiary history about the kidney disease.

I found this link useful in order to understand the difference b/w both of these. It looks they have RenalDisease indicator to represent whether the beneficiary has or had Kidney Failure. And, ChronicCond_KidneyDisease represents the long term Kidney Disease may be like not functioning to the fullest.

# Lets validate whether we have beneficiaries with both RKD & CKD?
with plt.style.context('seaborn'):
  plt.figure(figsize=(12,8))
  fig = train_bene_df.groupby(['RenalDiseaseIndicator','ChronicCond_KidneyDisease'])['Gender'].count().plot(kind='bar', color=['orange','green','purple','red'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.02), ha='center', fontsize=13.5)
  
  # Added the Description for one of the horizontal bar
  fig.annotate('This is interesting!!\n Beneficiaries with NO Long Term Kidney Malfunction but\n suffered from Kidney Failure.', 
               xy=(2.85, 0.85),  xycoords='data', xytext=(1.02, 0.65), textcoords='axes fraction', fontsize=12.5,
               arrowprops=dict(facecolor='black', shrink=0.03), horizontalalignment='right', verticalalignment='top')

  # Providing the labels and title to the graph
  plt.xlabel("\nRKD and CKD both are present?", fontdict=label_font_dict)
  plt.xticks(labels=[("RKD_No","CKD_Yes"), ("RKD_No","CKD_No"), ("RKD_Yes","CKD_Yes"), ("RKD_Yes","CKD_No")], ticks=[0,1,2,3], rotation=0, size=12)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Imbalance b/w males and females across the human races?\n", fontdict=title_font_dict)

# RKD --> Renal Disease Indicator
# CKD --> Chronic Condition Kidney Disease
print(pd.DataFrame(train_bene_df.groupby(['RenalDiseaseIndicator','ChronicCond_KidneyDisease'])['Gender'].count()),"\n")

                                                 Gender
RenalDiseaseIndicator ChronicCond_KidneyDisease        
0                     1                           27768
                      2                           91210
Y                     1                           15511
                      2                            4067 

• OBSERVATION
  • In the above, we came to know that 2.94% of beneficiaries had no previously CKD but suffered from RKD. It would be good to check how many out these end-up as FRAUD claims.

Q6. Lets see the number of beneficiaries on the basis of State Codes.

train_bene_df['State'].unique()

array([39, 52, 24, 23, 45, 15, 44, 41,  1, 14,  7, 13, 34, 31,  5, 49, 46,
        6, 38, 10, 26,  3, 20, 33, 18, 27, 51, 29, 42, 36, 11, 12, 30, 19,
        4, 25, 17, 16, 32,  2, 21, 50, 22, 54, 28, 43, 35, 37, 47,  8,  9,
       53], dtype=int64)

# Here, I'm displaying the distribution of BENEFICIARIES on the basis of State
with plt.style.context('seaborn'):
  plt.figure(figsize=(20,9))
  fig = train_bene_df['State'].value_counts().plot(kind='bar')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.03), ha='center', fontsize=13.5, rotation=90)
  # Providing the labels and title to the graph
  plt.xlabel("\nState Codes", fontdict=label_font_dict)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Distribution of BENEFICIARIES on the basis of States\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us that for some States the number of beneficiaries are more as compared to the others. And, the maximum number of beneficiaries are from State-5.

Q7. Lets see the number of beneficiaries on the basis of Country Codes.

# Here, I'm displaying the distribution of BENEFICIARIES on the basis of Country
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(18,8))
  fig = train_bene_df['County'].value_counts()[0:45].plot(kind='bar', color='palegreen')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.03), ha='center', fontsize=13.5, rotation=90)
  # Providing the labels and title to the graph
  plt.xlabel("\nCountry Codes", fontdict=label_font_dict)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Distribution of BENEFICIARIES on the basis of Countries (Top-45)", fontdict=title_font_dict)

# Countries with only handful of beneficiaries
train_bene_df['County'].value_counts()[::-1][0:45]

834    1
944    1
804    1
796    1
742    1
744    1
753    1
117    1
521    1
691    2
84     2
875    2
912    2
14     2
904    2
712    2
793    3
785    3
873    3
412    3
955    3
632    3
672    3
161    3
862    3
876    3
844    3
1      3
662    4
612    4
893    4
903    4
931    4
887    4
583    4
34     4
755    4
292    5
431    5
592    5
741    5
361    5
562    5
542    5
797    5
Name: County, dtype: int64

• OBSERVATION

  • The above graph shows us that distribution of beneficiaries based upon the country codes. Some having very high number of beneficiaries and others have very less.

• NOTE

  • One thing still needs to be evaluated that whether some specific state or country code has higher number of frauds?

Q8.1 Lets see the number of beneficiaries on the basis of 'NoOfMonths_PartACov'.

train_bene_df['NoOfMonths_PartACov'].unique()

array([12,  0,  3,  7, 10, 11,  8,  4,  1,  6,  2,  9,  5], dtype=int64)

# Here, I'm displaying the distribution of BENEFICIARIES on the basis of NoOfMonths_PartACov
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(18,8))
  fig = train_bene_df['NoOfMonths_PartACov'].value_counts().plot(kind='bar', color='palegreen')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.03), ha='center', fontsize=13.5, rotation=90)
  # Providing the labels and title to the graph
  plt.xlabel("\nMonths for Part-A Coverage", fontdict=label_font_dict)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Distribution of BENEFICIARIES on the basis of 'NoOfMonths_PartACov'", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us that more than 99% of beneficiaries have 12 months Part-A Coverage. My initial look to above plot tells me that this feature won't have much impact on the tgt labels.

Q8.2 Lets see the number of beneficiaries on the basis of 'NoOfMonths_PartBCov'.

train_bene_df['NoOfMonths_PartBCov'].unique()

array([12,  0,  1,  9, 11,  3, 10,  8,  2,  6,  4,  5,  7], dtype=int64)

# Here, I'm displaying the distribution of BENEFICIARIES on the basis of NoOfMonths_PartBCov
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(18,8))
  fig = train_bene_df['NoOfMonths_PartBCov'].value_counts().plot(kind='bar', color='palegreen')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.03), ha='center', fontsize=13.5, rotation=90)
  # Providing the labels and title to the graph
  plt.xlabel("\nMonths for Part-B Coverage", fontdict=label_font_dict)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Distribution of BENEFICIARIES on the basis of 'NoOfMonths_PartBCov'", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us that around 99% of beneficiaries have 12 months Part-B Coverage. My initial look to above plot tells me that this feature won't have much impact on the tgt labels.

Q9 Lets see the number of beneficiaries on the basis of 'ChronicCond_Alzheimer'. And, the Annual IP & OP expenditures for such patients.

# Here, I'm displaying the distribution of BENEFICIARIES on the basis of 'ChronicCond_Alzheimer'
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(12,8))
  fig = train_bene_df['ChronicCond_Alzheimer'].value_counts().plot(kind='bar', color=['palegreen','orange'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.01), ha='center', fontsize=13.5, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic ALZH Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['NO', 'YES'], fontsize=13, rotation=30)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Distribution of BENEFICIARIES on the basis of 'ChronicCond_Alzheimer'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic ALZH Disease
# 2 means -ve with Chronic ALZH Disease
print(pd.DataFrame(train_bene_df['ChronicCond_Alzheimer'].value_counts()),"\n")

   ChronicCond_Alzheimer
2                  92530
1                  46026 

• OBSERVATION
  • The above graph shows us that the beneficiaries with Chronic ALZH Disease are almost half as compared to the non-ALZH beneficiaries.

# Here, I'm displaying the Total Annual Sum of Max IP Reimbursement for 'ChronicCond_Alzheimer'
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(12,8))
  fig = train_bene_df.groupby(['ChronicCond_Alzheimer'])['IPAnnualReimbursementAmt'].sum().plot(kind='bar', color=['orange','palegreen'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/(train_bene_df["IPAnnualReimbursementAmt"].sum()),2))+"%"}', (x + width/2, y + height*1.01), ha='center', fontsize=13.5, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic ALZH Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum of Max IP Reimbursement \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of Max IP Reimbursement : 'ChronicCond_Alzheimer'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic ALZH Disease
# 2 means -ve with Chronic ALZH Disease
print(pd.DataFrame(train_bene_df.groupby(['ChronicCond_Alzheimer'])['IPAnnualReimbursementAmt'].sum()),"\n")

                       IPAnnualReimbursementAmt
ChronicCond_Alzheimer                          
1                                     247213280
2                                     259949690 

• OBSERVATION
  • The above graph shows us that despite the non-ALZH beneficiaries are double in number as compared to their counterparts, however, the Annual Sum of MAX IP Reimbursement is almost same for both the groups.
    • This means if a beneficiary with or without chronic ALZH Disease get admitted then the annual reimbursement paid is approx same.
      • The difference is of 13 Million or 2.52%.

# Here, I'm displaying the Total Annual Sum of Max OP Reimbursement for 'ChronicCond_Alzheimer'
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(12,8))
  fig = train_bene_df.groupby(['ChronicCond_Alzheimer'])['OPAnnualReimbursementAmt'].sum().plot(kind='bar', color=['orange','palegreen'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/(train_bene_df["OPAnnualReimbursementAmt"].sum()),2))+"%"}', (x + width/2, y + height*1.01), ha='center', fontsize=13.5, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic ALZH Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum of Max OP Reimbursement \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of Max OP Reimbursement : 'ChronicCond_Alzheimer'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic ALZH Disease
# 2 means -ve with Chronic ALZH Disease
print(pd.DataFrame(train_bene_df.groupby(['ChronicCond_Alzheimer'])['OPAnnualReimbursementAmt'].sum()),"\n")

                       OPAnnualReimbursementAmt
ChronicCond_Alzheimer                          
1                                      74721530
2                                     105154550 

• OBSERVATION
  • The above graph shows us that the Annual Sum of OP Reimbursement for non-chronic ALZH Disease beneficiaries is more than 31 Million or 16.92% than the other group.

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_Alzheimer'
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(12,8))
  fig = train_bene_df.groupby(['ChronicCond_Alzheimer'])['IPAnnualDeductibleAmt'].sum().plot(kind='bar', color=['orange','palegreen'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/(train_bene_df["IPAnnualDeductibleAmt"].sum()),2))+"%"}', (x + width/2, y + height*1.01), ha='center', fontsize=13.5, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic ALZH Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum of IP Co-payment \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of IP Co-payment paid by patient : 'ChronicCond_Alzheimer'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic ALZH Disease
# 2 means -ve with Chronic ALZH Disease
print(pd.DataFrame(train_bene_df.groupby(['ChronicCond_Alzheimer'])['IPAnnualDeductibleAmt'].sum()),"\n")

                       IPAnnualDeductibleAmt
ChronicCond_Alzheimer                       
1                                   28200854
2                                   27200388 

• OBSERVATION
  • The above graph shows us that the Annual Sum of IP Co-payment paid by the patient of non-chronic & chronic ALZH Disease is almost same. The difference is only of 10 Million or 1.8%.
    • So, we can say that the re-imbursement & co-payment amounts are same for chronic and non-chronic patients if they gets admitted.

# Here, I'm displaying the Total Annual Sum of OP Co-payment for 'ChronicCond_Alzheimer'
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(12,8))
  fig = train_bene_df.groupby(['ChronicCond_Alzheimer'])['OPAnnualDeductibleAmt'].sum().plot(kind='bar', color=['orange','palegreen'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/(train_bene_df["OPAnnualDeductibleAmt"].sum()),2))+"%"}', (x + width/2, y + height*1.01), ha='center', fontsize=13.5, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic ALZH Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum of OP Co-payment \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of OP Co-payment paid by patient : 'ChronicCond_Alzheimer'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic ALZH Disease
# 2 means -ve with Chronic ALZH Disease
print(pd.DataFrame(train_bene_df.groupby(['ChronicCond_Alzheimer'])['OPAnnualDeductibleAmt'].sum()),"\n")

                       OPAnnualDeductibleAmt
ChronicCond_Alzheimer                       
1                                   21502981
2                                   30832150 

• OBSERVATION
  • The above graph shows us that the Annual Sum of OP Co-payment paid by the patient of non-chronic is more as compared to chronic ALZH Disease. The difference is of 93.3 Million or 17.82%.
    • So, we can say that the co-payment paid by non-admitted patients is higher than the ones who gets admitted.

# Number of beneficiaries with chronic or no-chronic conditions
pd.DataFrame(train_bene_df.groupby(['ChronicCond_Alzheimer'])['BeneID'].count())

	BeneID
ChronicCond_Alzheimer
1	46026
2	92530

CC_ALZH_IP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Alzheimer'])['IPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_Alzheimer'])['BeneID'].count())
CC_ALZH_IP_R.columns = ['AVG IP Reimbursement Amt']
CC_ALZH_IP_R

	AVG IP Reimbursement Amt
ChronicCond_Alzheimer
1	5371.165863
2	2809.355777

CC_ALZH_OP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Alzheimer'])['OPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_Alzheimer'])['BeneID'].count())
CC_ALZH_OP_R.columns = ['AVG OP Reimbursement Amt']
CC_ALZH_OP_R

	AVG OP Reimbursement Amt
ChronicCond_Alzheimer
1	1623.463477
2	1136.437372

CC_ALZH_IP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Alzheimer'])['IPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_Alzheimer'])['BeneID'].count())
CC_ALZH_IP_D.columns = ['AVG IP Co-payment Amt']
CC_ALZH_IP_D

	AVG IP Co-payment Amt
ChronicCond_Alzheimer
1	612.715726
2	293.962909

CC_ALZH_OP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Alzheimer'])['OPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_Alzheimer'])['BeneID'].count())
CC_ALZH_OP_D.columns = ['AVG OP Co-payment Amt']
CC_ALZH_OP_D

	AVG OP Co-payment Amt
ChronicCond_Alzheimer
1	467.192044
2	333.212472

CC_ALZH_all_amts = pd.concat([CC_ALZH_IP_R, CC_ALZH_OP_R, CC_ALZH_IP_D, CC_ALZH_OP_D], axis=1)
CC_ALZH_all_amts

	AVG IP Reimbursement Amt	AVG OP Reimbursement Amt	AVG IP Co-payment Amt	AVG OP Co-payment Amt
ChronicCond_Alzheimer
1	5371.165863	1623.463477	612.715726	467.192044
2	2809.355777	1136.437372	293.962909	333.212472

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_Alzheimer'
with plt.style.context('seaborn-poster'):
  fig = CC_ALZH_all_amts.plot(kind='bar', colormap='rainbow')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{round(height,0)}', (x + width/2, y + height*1.015), ha='center', fontsize=12, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic ALZH Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="-.", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of various amounts : 'ChronicCond_Alzheimer'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic ALZH Disease
# 2 means -ve with Chronic ALZH Disease
print(CC_ALZH_all_amts,"\n")

                       AVG IP Reimbursement Amt  AVG OP Reimbursement Amt  \
ChronicCond_Alzheimer                                                       
1                                   5371.165863               1623.463477   
2                                   2809.355777               1136.437372   

                       AVG IP Co-payment Amt  AVG OP Co-payment Amt  
ChronicCond_Alzheimer                                                
1                                 612.715726             467.192044  
2                                 293.962909             333.212472   

• OBSERVATION
  • The above graph is telling us below points:
    • Payer pays huge chunk of expenses specially when a beneficiary gets admitted with or without chronic ALZH disease. To be more precise, in case of +ve it is more than 50%.
    • For other comparisons the difference is not very high.

Q10. Lets see the number of beneficiaries on the basis of 'ChronicCond_Heartfailure'. And, the Annual IP & OP expenditures for such patients.

# Here, I'm displaying the distribution of BENEFICIARIES on the basis of 'ChronicCond_Heartfailure'
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(12,8))
  fig = train_bene_df['ChronicCond_Heartfailure'].value_counts().plot(kind='bar', color=['palegreen','orange'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/train_bene_df.shape[0],2))+"%"}', (x + width/2, y + height*1.01), ha='center', fontsize=13.5, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic HF Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['NO', 'YES'], fontsize=13, rotation=30)
  plt.ylabel("Number or % share of patients\n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Distribution of BENEFICIARIES on the basis of 'ChronicCond_Heartfailure'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic HF Disease
# 2 means -ve with Chronic HF Disease
print(pd.DataFrame(train_bene_df['ChronicCond_Heartfailure'].value_counts()),"\n")

   ChronicCond_Heartfailure
2                     70154
1                     68402 

• OBSERVATION
  • The above graph shows us that the beneficiaries with Chronic HF Disease are almost equal to the non-ALZH beneficiaries.

# Here, I'm displaying the Total Annual Sum of Max IP Reimbursement for 'ChronicCond_Heartfailure'
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(12,8))
  fig = train_bene_df.groupby(['ChronicCond_Heartfailure'])['IPAnnualReimbursementAmt'].sum().plot(kind='bar', color=['orange','palegreen'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/(train_bene_df["IPAnnualReimbursementAmt"].sum()),2))+"%"}', (x + width/2, y + height*1.01), ha='center', fontsize=13.5, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic HF Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum of Max IP Reimbursement \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of Max IP Reimbursement : 'ChronicCond_Heartfailure'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic HF Disease
# 2 means -ve with Chronic HF Disease
print(pd.DataFrame(train_bene_df.groupby(['ChronicCond_Heartfailure'])['IPAnnualReimbursementAmt'].sum()),"\n")

                          IPAnnualReimbursementAmt
ChronicCond_Heartfailure                          
1                                        370856140
2                                        136306830 

• OBSERVATION
  • The above graph shows us that the Payer pays very high amount annually for patients with Chronic Heart Failure Disease.
    • The difference is of around 234 Million or 47%. This is very high and it can be a potential sign of FRAUD claims.

# Here, I'm displaying the Total Annual Sum of Max OP Reimbursement for 'ChronicCond_Heartfailure'
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(12,8))
  fig = train_bene_df.groupby(['ChronicCond_Heartfailure'])['OPAnnualReimbursementAmt'].sum().plot(kind='bar', color=['orange','palegreen'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/(train_bene_df["OPAnnualReimbursementAmt"].sum()),2))+"%"}', (x + width/2, y + height*1.01), ha='center', fontsize=13.5, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic HF Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum of Max OP Reimbursement \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of Max OP Reimbursement : 'ChronicCond_Heartfailure'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic HF Disease
# 2 means -ve with Chronic HF Disease
print(pd.DataFrame(train_bene_df.groupby(['ChronicCond_Heartfailure'])['OPAnnualReimbursementAmt'].sum()),"\n")

                          OPAnnualReimbursementAmt
ChronicCond_Heartfailure                          
1                                        114280500
2                                         65595580 

• OBSERVATION
  • The above graph shows us that the Annual Sum paid by the payers for non-admitted patients is 49 Million more than the other group. This again can be a potential sign of fraud claims.

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_Heartfailure'
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(12,8))
  fig = train_bene_df.groupby(['ChronicCond_Heartfailure'])['IPAnnualDeductibleAmt'].sum().plot(kind='bar', color=['orange','palegreen'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/(train_bene_df["IPAnnualDeductibleAmt"].sum()),2))+"%"}', (x + width/2, y + height*1.01), ha='center', fontsize=13.5, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic HF Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum of IP Co-payment \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of IP Co-payment paid by patient : 'ChronicCond_Heartfailure'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic HF Disease
# 2 means -ve with Chronic HF Disease
print(pd.DataFrame(train_bene_df.groupby(['ChronicCond_Heartfailure'])['IPAnnualDeductibleAmt'].sum()),"\n")

                          IPAnnualDeductibleAmt
ChronicCond_Heartfailure                       
1                                      40303650
2                                      15097592 

• OBSERVATION
  • The above graph shows us that the Annual Sum of IP Co-payment paid by the patient of non-chronic & chronic HF Disease has a difference of 25 Million.
    • But, the point to note here that the co-payment proportion is very less in front of the part paid by the Payer. This again can be potential sign of FRAUD.

# Here, I'm displaying the Total Annual Sum of OP Co-payment for 'ChronicCond_Heartfailure'
with plt.style.context('seaborn-poster'):
  plt.figure(figsize=(12,8))
  fig = train_bene_df.groupby(['ChronicCond_Heartfailure'])['OPAnnualDeductibleAmt'].sum().plot(kind='bar', color=['orange','palegreen'])
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{str(round((height*100)/(train_bene_df["OPAnnualDeductibleAmt"].sum()),2))+"%"}', (x + width/2, y + height*1.01), ha='center', fontsize=13.5, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic HF Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum of OP Co-payment \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="--", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of OP Co-payment paid by patient : 'ChronicCond_Heartfailure'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic HF Disease
# 2 means -ve with Chronic HF Disease
print(pd.DataFrame(train_bene_df.groupby(['ChronicCond_Heartfailure'])['OPAnnualDeductibleAmt'].sum()),"\n")

                          OPAnnualDeductibleAmt
ChronicCond_Heartfailure                       
1                                      32649838
2                                      19685293 

• OBSERVATION
  • The above graph shows us that the Annual Sum of OP Co-payment paid by the patient of non-chronic & chronic HF Disease has a difference of 13 Million.
    • But, the point to note here that the co-payment proportion is very less in front of the part paid by the Payer. This again can be potential sign of FRAUD.

# Number of beneficiaries with chronic or no-chronic conditions
pd.DataFrame(train_bene_df.groupby(['ChronicCond_Heartfailure'])['BeneID'].count())

	BeneID
ChronicCond_Heartfailure
1	68402
2	70154

CC_HF_IP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Heartfailure'])['IPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_Heartfailure'])['BeneID'].count())
CC_HF_IP_R.columns = ['AVG IP Reimbursement Amt']
CC_HF_IP_R

	AVG IP Reimbursement Amt
ChronicCond_Heartfailure
1	5421.714862
2	1942.965904

CC_HF_OP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Heartfailure'])['OPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_Heartfailure'])['BeneID'].count())
CC_HF_OP_R.columns = ['AVG OP Reimbursement Amt']
CC_HF_OP_R

	AVG OP Reimbursement Amt
ChronicCond_Heartfailure
1	1670.718692
2	935.022664

CC_HF_IP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Heartfailure'])['IPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_Heartfailure'])['BeneID'].count())
CC_HF_IP_D.columns = ['AVG IP Co-payment Amt']
CC_HF_IP_D

	AVG IP Co-payment Amt
ChronicCond_Heartfailure
1	589.217421
2	215.206432

CC_HF_OP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Heartfailure'])['OPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_Heartfailure'])['BeneID'].count())
CC_HF_OP_D.columns = ['AVG OP Co-payment Amt']
CC_HF_OP_D

	AVG OP Co-payment Amt
ChronicCond_Heartfailure
1	477.322856
2	280.601149

CC_HF_all_amts = pd.concat([CC_HF_IP_R, CC_HF_OP_R, CC_HF_IP_D, CC_HF_OP_D], axis=1)
CC_HF_all_amts

	AVG IP Reimbursement Amt	AVG OP Reimbursement Amt	AVG IP Co-payment Amt	AVG OP Co-payment Amt
ChronicCond_Heartfailure
1	5421.714862	1670.718692	589.217421	477.322856
2	1942.965904	935.022664	215.206432	280.601149

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_Heartfailure'
with plt.style.context('seaborn-poster'):
  fig = CC_HF_all_amts.plot(kind='bar', colormap='rainbow')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{round(height,0)}', (x + width/2, y + height*1.015), ha='center', fontsize=12, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic HF Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="-.", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of various amounts : 'ChronicCond_Heartfailure'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic HF Disease
# 2 means -ve with Chronic HF Disease
print(CC_HF_all_amts,"\n")

                          AVG IP Reimbursement Amt  AVG OP Reimbursement Amt  \
ChronicCond_Heartfailure                                                       
1                                      5421.714862               1670.718692   
2                                      1942.965904                935.022664   

                          AVG IP Co-payment Amt  AVG OP Co-payment Amt  
ChronicCond_Heartfailure                                                
1                                    589.217421             477.322856  
2                                    215.206432             280.601149   

• OBSERVATION
  • The above graph is telling us below points:
    • Payer pays huge chunk of expenses specially when a beneficiary gets admitted with or without chronic HF disease. To be more precise, in case of +ve it is more than 50%.
    • For other comparisons the difference is not very high.

Q11. Lets see the number of beneficiaries on the basis of 'ChronicCond_KidneyDisease'. And, the Annual IP & OP expenditures for such patients.

# Number of beneficiaries with chronic or no-chronic conditions
pd.DataFrame(train_bene_df.groupby(['ChronicCond_KidneyDisease'])['BeneID'].count())

	BeneID
ChronicCond_KidneyDisease
1	43279
2	95277

CC_KD_IP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_KidneyDisease'])['IPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_KidneyDisease'])['BeneID'].count())
CC_KD_IP_R.columns = ['AVG IP Reimbursement Amt']
CC_KD_IP_R

	AVG IP Reimbursement Amt
ChronicCond_KidneyDisease
1	7500.768040
2	1915.858287

CC_KD_OP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_KidneyDisease'])['OPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_KidneyDisease'])['BeneID'].count())
CC_KD_OP_R.columns = ['AVG OP Reimbursement Amt']
CC_KD_OP_R

	AVG OP Reimbursement Amt
ChronicCond_KidneyDisease
1	2061.602625
2	951.457120

CC_KD_IP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_KidneyDisease'])['IPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_KidneyDisease'])['BeneID'].count())
CC_KD_IP_D.columns = ['AVG IP Co-payment Amt']
CC_KD_IP_D

	AVG IP Co-payment Amt
ChronicCond_KidneyDisease
1	785.385707
2	224.718809

CC_KD_OP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_KidneyDisease'])['OPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_KidneyDisease'])['BeneID'].count())
CC_KD_OP_D.columns = ['AVG OP Co-payment Amt']
CC_KD_OP_D

	AVG OP Co-payment Amt
ChronicCond_KidneyDisease
1	582.706417
2	284.603629

CC_KD_all_amts = pd.concat([CC_KD_IP_R, CC_KD_OP_R, CC_KD_IP_D, CC_KD_OP_D], axis=1)
CC_KD_all_amts

	AVG IP Reimbursement Amt	AVG OP Reimbursement Amt	AVG IP Co-payment Amt	AVG OP Co-payment Amt
ChronicCond_KidneyDisease
1	7500.768040	2061.602625	785.385707	582.706417
2	1915.858287	951.457120	224.718809	284.603629

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_KidneyDisease'
with plt.style.context('seaborn-poster'):
  fig = CC_KD_all_amts.plot(kind='bar', colormap='rainbow')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{round(height,0)}', (x + width/2, y + height*1.015), ha='center', fontsize=12, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic KD Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="-.", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of various amounts : 'ChronicCond_KidneyDisease'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic KD Disease
# 2 means -ve with Chronic KD Disease
print(CC_KD_all_amts,"\n")

                           AVG IP Reimbursement Amt  AVG OP Reimbursement Amt  \
ChronicCond_KidneyDisease                                                       
1                                       7500.768040               2061.602625   
2                                       1915.858287                951.457120   

                           AVG IP Co-payment Amt  AVG OP Co-payment Amt  
ChronicCond_KidneyDisease                                                
1                                     785.385707             582.706417  
2                                     224.718809             284.603629   

• OBSERVATION
  • The above graph is telling us below points:
    • Payer pays huge chunk of expenses specially when a beneficiary gets admitted with or without chronic HF disease. To be more precise, in case of +ve it is more than 50%.
    • For other comparisons the difference is not very high.

Q12. Lets see the number of beneficiaries on the basis of 'ChronicCond_Cancer'. And, the Annual IP & OP expenditures for such patients.

# Number of beneficiaries with chronic or no-chronic conditions
pd.DataFrame(train_bene_df.groupby(['ChronicCond_Cancer'])['BeneID'].count())

	BeneID
ChronicCond_Cancer
1	16621
2	121935

CC_CN_IP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Cancer'])['IPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_Cancer'])['BeneID'].count())
CC_CN_IP_R.columns = ['AVG IP Reimbursement Amt']
CC_CN_IP_R

	AVG IP Reimbursement Amt
ChronicCond_Cancer
1	6067.532038
2	3332.222250

CC_CN_OP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Cancer'])['OPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_Cancer'])['BeneID'].count())
CC_CN_OP_R.columns = ['AVG OP Reimbursement Amt']
CC_CN_OP_R

	AVG OP Reimbursement Amt
ChronicCond_Cancer
1	1788.427291
2	1231.398942

CC_CN_IP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Cancer'])['IPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_Cancer'])['BeneID'].count())
CC_CN_IP_D.columns = ['AVG IP Co-payment Amt']
CC_CN_IP_D

	AVG IP Co-payment Amt
ChronicCond_Cancer
1	636.339811
2	367.610924

CC_CN_OP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Cancer'])['OPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_Cancer'])['BeneID'].count())
CC_CN_OP_D.columns = ['AVG OP Co-payment Amt']
CC_CN_OP_D

	AVG OP Co-payment Amt
ChronicCond_Cancer
1	520.832682
2	358.210284

CC_CN_all_amts = pd.concat([CC_CN_IP_R, CC_CN_OP_R, CC_CN_IP_D, CC_CN_OP_D], axis=1)
CC_CN_all_amts

	AVG IP Reimbursement Amt	AVG OP Reimbursement Amt	AVG IP Co-payment Amt	AVG OP Co-payment Amt
ChronicCond_Cancer
1	6067.532038	1788.427291	636.339811	520.832682
2	3332.222250	1231.398942	367.610924	358.210284

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_Cancer'
with plt.style.context('seaborn-poster'):
  fig = CC_CN_all_amts.plot(kind='bar', colormap='rainbow')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{round(height,0)}', (x + width/2, y + height*1.015), ha='center', fontsize=12, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic CN Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="-.", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of various amounts : 'ChronicCond_Cancer'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic CN Disease
# 2 means -ve with Chronic CN Disease
print(CC_CN_all_amts,"\n")

                    AVG IP Reimbursement Amt  AVG OP Reimbursement Amt  \
ChronicCond_Cancer                                                       
1                                6067.532038               1788.427291   
2                                3332.222250               1231.398942   

                    AVG IP Co-payment Amt  AVG OP Co-payment Amt  
ChronicCond_Cancer                                                
1                              636.339811             520.832682  
2                              367.610924             358.210284   

• OBSERVATION
  • The above graph is telling us below points:
    • Payer pays huge chunk of expenses specially when a beneficiary gets admitted with or without chronic HF disease. To be more precise, in case of +ve it is around 50%.
    • For other comparisons the difference is not very high.

Q13. Lets see the number of beneficiaries on the basis of 'ChronicCond_ObstrPulmonary'. And, the Annual IP & OP expenditures for such patients.

# Number of beneficiaries with chronic or no-chronic conditions
pd.DataFrame(train_bene_df.groupby(['ChronicCond_ObstrPulmonary'])['BeneID'].count())

	BeneID
ChronicCond_ObstrPulmonary
1	32859
2	105697

CC_PL_IP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_ObstrPulmonary'])['IPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_ObstrPulmonary'])['BeneID'].count())
CC_PL_IP_R.columns = ['AVG IP Reimbursement Amt']
CC_PL_IP_R

	AVG IP Reimbursement Amt
ChronicCond_ObstrPulmonary
1	7361.772422
2	2509.650132

CC_PL_OP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_ObstrPulmonary'])['OPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_ObstrPulmonary'])['BeneID'].count())
CC_PL_OP_R.columns = ['AVG OP Reimbursement Amt']
CC_PL_OP_R

	AVG OP Reimbursement Amt
ChronicCond_ObstrPulmonary
1	1810.864908
2	1138.848501

CC_PL_IP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_ObstrPulmonary'])['IPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_ObstrPulmonary'])['BeneID'].count())
CC_PL_IP_D.columns = ['AVG IP Co-payment Amt']
CC_PL_IP_D

	AVG IP Co-payment Amt
ChronicCond_ObstrPulmonary
1	818.228917
2	269.781148

CC_PL_OP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_ObstrPulmonary'])['OPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_ObstrPulmonary'])['BeneID'].count())
CC_PL_OP_D.columns = ['AVG OP Co-payment Amt']
CC_PL_OP_D

	AVG OP Co-payment Amt
ChronicCond_ObstrPulmonary
1	519.44767
2	333.65753

CC_PL_all_amts = pd.concat([CC_PL_IP_R, CC_PL_OP_R, CC_PL_IP_D, CC_PL_OP_D], axis=1)
CC_PL_all_amts

	AVG IP Reimbursement Amt	AVG OP Reimbursement Amt	AVG IP Co-payment Amt	AVG OP Co-payment Amt
ChronicCond_ObstrPulmonary
1	7361.772422	1810.864908	818.228917	519.44767
2	2509.650132	1138.848501	269.781148	333.65753

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_ObstrPulmonary'
with plt.style.context('seaborn-poster'):
  fig = CC_PL_all_amts.plot(kind='bar', colormap='rainbow')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{round(height,0)}', (x + width/2, y + height*1.015), ha='center', fontsize=12, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic PL Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="-.", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of various amounts : 'ChronicCond_ObstrPulmonary'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic PL Disease
# 2 means -ve with Chronic PL Disease

• OBSERVATION
  • The above graph is telling us below points:
    • Payer pays huge chunk of expenses specially when a beneficiary gets admitted with or without chronic HF disease. To be more precise, in case of +ve it is more than 50%.
    • For other comparisons the difference is not very high.

Q14. Lets see the number of beneficiaries on the basis of 'ChronicCond_Depression'. And, the Annual IP & OP expenditures for such patients.

# Number of beneficiaries with chronic or no-chronic conditions
pd.DataFrame(train_bene_df.groupby(['ChronicCond_Depression'])['BeneID'].count())

	BeneID
ChronicCond_Depression
1	49260
2	89296

CC_DP_IP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Depression'])['IPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_Depression'])['BeneID'].count())
CC_DP_IP_R.columns = ['AVG IP Reimbursement Amt']
CC_DP_IP_R

	AVG IP Reimbursement Amt
ChronicCond_Depression
1	5022.162810
2	2909.102647

CC_DP_OP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Depression'])['OPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_Depression'])['BeneID'].count())
CC_DP_OP_R.columns = ['AVG OP Reimbursement Amt']
CC_DP_OP_R

	AVG OP Reimbursement Amt
ChronicCond_Depression
1	1603.871701
2	1129.606701

CC_DP_IP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Depression'])['IPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_Depression'])['BeneID'].count())
CC_DP_IP_D.columns = ['AVG IP Co-payment Amt']
CC_DP_IP_D

	AVG IP Co-payment Amt
ChronicCond_Depression
1	570.054771
2	305.952607

CC_DP_OP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Depression'])['OPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_Depression'])['BeneID'].count())
CC_DP_OP_D.columns = ['AVG OP Co-payment Amt']
CC_DP_OP_D

	AVG OP Co-payment Amt
ChronicCond_Depression
1	464.285708
2	329.963459

CC_DP_all_amts = pd.concat([CC_DP_IP_R, CC_DP_OP_R, CC_DP_IP_D, CC_DP_OP_D], axis=1)
CC_DP_all_amts

	AVG IP Reimbursement Amt	AVG OP Reimbursement Amt	AVG IP Co-payment Amt	AVG OP Co-payment Amt
ChronicCond_Depression
1	5022.162810	1603.871701	570.054771	464.285708
2	2909.102647	1129.606701	305.952607	329.963459

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_Depression'
with plt.style.context('seaborn-poster'):
  fig = CC_DP_all_amts.plot(kind='bar', colormap='rainbow')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{round(height,0)}', (x + width/2, y + height*1.015), ha='center', fontsize=12, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic DP Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="-.", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of various amounts : 'ChronicCond_Depression'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic DP Disease
# 2 means -ve with Chronic DP Disease

• OBSERVATION
  • The above graph is telling us below points:
    • Payer pays huge chunk of expenses specially when a beneficiary gets admitted with or without chronic HF disease. To be more precise, in case of +ve it is around 42%.
    • For other comparisons the difference is not very high.

Q15. Lets see the number of beneficiaries on the basis of 'ChronicCond_Diabetes'. And, the Annual IP & OP expenditures for such patients.

# Number of beneficiaries with chronic or no-chronic conditions
pd.DataFrame(train_bene_df.groupby(['ChronicCond_Diabetes'])['BeneID'].count())

	BeneID
ChronicCond_Diabetes
1	83391
2	55165

CC_DB_IP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Diabetes'])['IPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_Diabetes'])['BeneID'].count())
CC_DB_IP_R.columns = ['AVG IP Reimbursement Amt']
CC_DB_IP_R

	AVG IP Reimbursement Amt
ChronicCond_Diabetes
1	4870.622010
2	1830.815372

CC_DB_OP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Diabetes'])['OPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_Diabetes'])['BeneID'].count())
CC_DB_OP_R.columns = ['AVG OP Reimbursement Amt']
CC_DB_OP_R

	AVG OP Reimbursement Amt
ChronicCond_Diabetes
1	1607.259536
2	831.054110

CC_DB_IP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Diabetes'])['IPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_Diabetes'])['BeneID'].count())
CC_DB_IP_D.columns = ['AVG IP Co-payment Amt']
CC_DB_IP_D

	AVG IP Co-payment Amt
ChronicCond_Diabetes
1	534.254680
2	196.668359

CC_DB_OP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Diabetes'])['OPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_Diabetes'])['BeneID'].count())
CC_DB_OP_D.columns = ['AVG OP Co-payment Amt']
CC_DB_OP_D

	AVG OP Co-payment Amt
ChronicCond_Diabetes
1	464.345421
2	246.767008

CC_DB_all_amts = pd.concat([CC_DB_IP_R, CC_DB_OP_R, CC_DB_IP_D, CC_DB_OP_D], axis=1)
CC_DB_all_amts

	AVG IP Reimbursement Amt	AVG OP Reimbursement Amt	AVG IP Co-payment Amt	AVG OP Co-payment Amt
ChronicCond_Diabetes
1	4870.622010	1607.259536	534.254680	464.345421
2	1830.815372	831.054110	196.668359	246.767008

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_Diabetes'
with plt.style.context('seaborn-poster'):
  fig = CC_DB_all_amts.plot(kind='bar', colormap='rainbow')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{round(height,0)}', (x + width/2, y + height*1.015), ha='center', fontsize=12, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic DB Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="-.", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of various amounts : 'ChronicCond_Diabetes'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic DB Disease
# 2 means -ve with Chronic DB Disease

• OBSERVATION
  • The above graph is telling us below points:
    • Payer pays huge chunk of expenses specially when a beneficiary gets admitted with or without chronic HF disease. To be more precise, in case of +ve it is more than 50%.
    • For other comparisons the difference is not very high.

Q16. Lets see the number of beneficiaries on the basis of 'ChronicCond_IschemicHeart'. And, the Annual IP & OP expenditures for such patients.

# Number of beneficiaries with chronic or no-chronic conditions
pd.DataFrame(train_bene_df.groupby(['ChronicCond_IschemicHeart'])['BeneID'].count())

	BeneID
ChronicCond_IschemicHeart
1	93644
2	44912

CC_IH_IP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_IschemicHeart'])['IPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_IschemicHeart'])['BeneID'].count())
CC_IH_IP_R.columns = ['AVG IP Reimbursement Amt']
CC_IH_IP_R

	AVG IP Reimbursement Amt
ChronicCond_IschemicHeart
1	4697.683034
2	1497.442554

CC_IH_OP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_IschemicHeart'])['OPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_IschemicHeart'])['BeneID'].count())
CC_IH_OP_R.columns = ['AVG OP Reimbursement Amt']
CC_IH_OP_R

	AVG OP Reimbursement Amt
ChronicCond_IschemicHeart
1	1521.077485
2	833.547827

CC_IH_IP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_IschemicHeart'])['IPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_IschemicHeart'])['BeneID'].count())
CC_IH_IP_D.columns = ['AVG IP Co-payment Amt']
CC_IH_IP_D

	AVG IP Co-payment Amt
ChronicCond_IschemicHeart
1	506.661676
2	177.133416

CC_IH_OP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_IschemicHeart'])['OPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_IschemicHeart'])['BeneID'].count())
CC_IH_OP_D.columns = ['AVG OP Co-payment Amt']
CC_IH_OP_D

	AVG OP Co-payment Amt
ChronicCond_IschemicHeart
1	441.447535
2	244.839197

CC_IH_all_amts = pd.concat([CC_IH_IP_R, CC_IH_OP_R, CC_IH_IP_D, CC_IH_OP_D], axis=1)
CC_IH_all_amts

	AVG IP Reimbursement Amt	AVG OP Reimbursement Amt	AVG IP Co-payment Amt	AVG OP Co-payment Amt
ChronicCond_IschemicHeart
1	4697.683034	1521.077485	506.661676	441.447535
2	1497.442554	833.547827	177.133416	244.839197

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_IschemicHeart'
with plt.style.context('seaborn-poster'):
  fig = CC_IH_all_amts.plot(kind='bar', colormap='rainbow')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{round(height,0)}', (x + width/2, y + height*1.015), ha='center', fontsize=12, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic IH Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="-.", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of various amounts : 'ChronicCond_IschemicHeart'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic IH Disease
# 2 means -ve with Chronic IH Disease

• OBSERVATION
  • The above graph is telling us below points:
    • Payer pays huge chunk of expenses specially when a beneficiary gets admitted with or without chronic HF disease. To be more precise, in case of +ve it is more than 50%.
    • For other comparisons the difference is not very high.

Q17. Lets see the number of beneficiaries on the basis of 'ChronicCond_Osteoporasis'. And, the Annual IP & OP expenditures for such patients.

# Number of beneficiaries with chronic or no-chronic conditions
pd.DataFrame(train_bene_df.groupby(['ChronicCond_Osteoporasis'])['BeneID'].count())

	BeneID
ChronicCond_Osteoporasis
1	38059
2	100497

CC_OS_IP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Osteoporasis'])['IPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_Osteoporasis'])['BeneID'].count())
CC_OS_IP_R.columns = ['AVG IP Reimbursement Amt']
CC_OS_IP_R

	AVG IP Reimbursement Amt
ChronicCond_Osteoporasis
1	4608.419822
2	3301.303721

CC_OS_OP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Osteoporasis'])['OPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_Osteoporasis'])['BeneID'].count())
CC_OS_OP_R.columns = ['AVG OP Reimbursement Amt']
CC_OS_OP_R

	AVG OP Reimbursement Amt
ChronicCond_Osteoporasis
1	1494.709004
2	1223.807178

CC_OS_IP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Osteoporasis'])['IPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_Osteoporasis'])['BeneID'].count())
CC_OS_IP_D.columns = ['AVG IP Co-payment Amt']
CC_OS_IP_D

	AVG IP Co-payment Amt
ChronicCond_Osteoporasis
1	512.038204
2	357.359722

CC_OS_OP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_Osteoporasis'])['OPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_Osteoporasis'])['BeneID'].count())
CC_OS_OP_D.columns = ['AVG OP Co-payment Amt']
CC_OS_OP_D

	AVG OP Co-payment Amt
ChronicCond_Osteoporasis
1	436.249271
2	355.552106

CC_OS_all_amts = pd.concat([CC_OS_IP_R, CC_OS_OP_R, CC_OS_IP_D, CC_OS_OP_D], axis=1)
CC_OS_all_amts

	AVG IP Reimbursement Amt	AVG OP Reimbursement Amt	AVG IP Co-payment Amt	AVG OP Co-payment Amt
ChronicCond_Osteoporasis
1	4608.419822	1494.709004	512.038204	436.249271
2	3301.303721	1223.807178	357.359722	355.552106

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_Osteoporasis'
with plt.style.context('seaborn-poster'):
  fig = CC_OS_all_amts.plot(kind='bar', colormap='rainbow')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{round(height,0)}', (x + width/2, y + height*1.015), ha='center', fontsize=12, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic OS Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="-.", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of various amounts : 'ChronicCond_Osteoporasis'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic OS Disease
# 2 means -ve with Chronic OS Disease

• OBSERVATION
  • The above graph is telling us below points:
    • Payer pays huge chunk of expenses specially when a beneficiary gets admitted with or without chronic HF disease. To be more precise, in case of +ve the difference is not that high.
    • For other comparisons the difference is not very high.

Q18. Lets see the number of beneficiaries on the basis of 'ChronicCond_rheumatoidarthritis'. And, the Annual IP & OP expenditures for such patients.

# Number of beneficiaries with chronic or no-chronic conditions
pd.DataFrame(train_bene_df.groupby(['ChronicCond_rheumatoidarthritis'])['BeneID'].count())

	BeneID
ChronicCond_rheumatoidarthritis
1	35584
2	102972

CC_RH_IP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_rheumatoidarthritis'])['IPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_rheumatoidarthritis'])['BeneID'].count())
CC_RH_IP_R.columns = ['AVG IP Reimbursement Amt']
CC_RH_IP_R

	AVG IP Reimbursement Amt
ChronicCond_rheumatoidarthritis
1	5102.027597
2	3162.145243

CC_RH_OP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_rheumatoidarthritis'])['OPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_rheumatoidarthritis'])['BeneID'].count())
CC_RH_OP_R.columns = ['AVG OP Reimbursement Amt']
CC_RH_OP_R

	AVG OP Reimbursement Amt
ChronicCond_rheumatoidarthritis
1	1548.057835
2	1211.882745

CC_RH_IP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_rheumatoidarthritis'])['IPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_rheumatoidarthritis'])['BeneID'].count())
CC_RH_IP_D.columns = ['AVG IP Co-payment Amt']
CC_RH_IP_D

	AVG IP Co-payment Amt
ChronicCond_rheumatoidarthritis
1	562.630789
2	343.594259

CC_RH_OP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_rheumatoidarthritis'])['OPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_rheumatoidarthritis'])['BeneID'].count())
CC_RH_OP_D.columns = ['AVG OP Co-payment Amt']
CC_RH_OP_D

	AVG OP Co-payment Amt
ChronicCond_rheumatoidarthritis
1	453.423196
2	351.556928

CC_RH_all_amts = pd.concat([CC_RH_IP_R, CC_RH_OP_R, CC_RH_IP_D, CC_RH_OP_D], axis=1)
CC_RH_all_amts

	AVG IP Reimbursement Amt	AVG OP Reimbursement Amt	AVG IP Co-payment Amt	AVG OP Co-payment Amt
ChronicCond_rheumatoidarthritis
1	5102.027597	1548.057835	562.630789	453.423196
2	3162.145243	1211.882745	343.594259	351.556928

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_rheumatoidarthritis'
with plt.style.context('seaborn-poster'):
  fig = CC_RH_all_amts.plot(kind='bar', colormap='rainbow')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{round(height,0)}', (x + width/2, y + height*1.015), ha='center', fontsize=12, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic RH Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="-.", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of various amounts : 'ChronicCond_rheumatoidarthritis'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic RH Disease
# 2 means -ve with Chronic RH Disease

• OBSERVATION
  • The above graph is telling us below points:
    • Payer pays huge chunk of expenses specially when a beneficiary gets admitted with or without chronic HF disease. To be more precise, in case of +ve the difference is not that high.
    • For other comparisons the difference is not very high.

Q19. Lets see the number of beneficiaries on the basis of 'ChronicCond_stroke'. And, the Annual IP & OP expenditures for such patients.

# Number of beneficiaries with chronic or no-chronic conditions
pd.DataFrame(train_bene_df.groupby(['ChronicCond_stroke'])['BeneID'].count())

	BeneID
ChronicCond_stroke
1	10954
2	127602

CC_ST_IP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_stroke'])['IPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_stroke'])['BeneID'].count())
CC_ST_IP_R.columns = ['AVG IP Reimbursement Amt']
CC_ST_IP_R

	AVG IP Reimbursement Amt
ChronicCond_stroke
1	8111.475260
2	3278.239134

CC_ST_OP_R = pd.DataFrame(train_bene_df.groupby(['ChronicCond_stroke'])['OPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['ChronicCond_stroke'])['BeneID'].count())
CC_ST_OP_R.columns = ['AVG OP Reimbursement Amt']
CC_ST_OP_R

	AVG OP Reimbursement Amt
ChronicCond_stroke
1	1925.197188
2	1244.396404

CC_ST_IP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_stroke'])['IPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_stroke'])['BeneID'].count())
CC_ST_IP_D.columns = ['AVG IP Co-payment Amt']
CC_ST_IP_D

	AVG IP Co-payment Amt
ChronicCond_stroke
1	861.399489
2	360.225326

CC_ST_OP_D = pd.DataFrame(train_bene_df.groupby(['ChronicCond_stroke'])['OPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['ChronicCond_stroke'])['BeneID'].count())
CC_ST_OP_D.columns = ['AVG OP Co-payment Amt']
CC_ST_OP_D

	AVG OP Co-payment Amt
ChronicCond_stroke
1	535.947508
2	364.135061

CC_ST_all_amts = pd.concat([CC_ST_IP_R, CC_ST_OP_R, CC_ST_IP_D, CC_ST_OP_D], axis=1)
CC_ST_all_amts

	AVG IP Reimbursement Amt	AVG OP Reimbursement Amt	AVG IP Co-payment Amt	AVG OP Co-payment Amt
ChronicCond_stroke
1	8111.475260	1925.197188	861.399489	535.947508
2	3278.239134	1244.396404	360.225326	364.135061

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'ChronicCond_stroke'
with plt.style.context('seaborn-poster'):
  fig = CC_ST_all_amts.plot(kind='bar', colormap='rainbow')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{round(height,0)}', (x + width/2, y + height*1.015), ha='center', fontsize=12, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic ST Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['YES', 'NO'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="-.", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of various amounts : 'ChronicCond_stroke'\n", fontdict=title_font_dict)

# 1 means +ve with Chronic ST Disease
# 2 means -ve with Chronic ST Disease

• OBSERVATION
  • The above graph is telling us below points:
    • Payer pays huge chunk of expenses specially when a beneficiary gets admitted with or without chronic HF disease. To be more precise, in case of +ve the difference is more than 50%.
    • For other comparisons the difference is not very high.

Q20. Lets see the number of beneficiaries on the basis of 'RenalDiseaseIndicator'. And, the Annual IP & OP expenditures for such patients.

# Number of beneficiaries with chronic or no-chronic conditions
pd.DataFrame(train_bene_df.groupby(['RenalDiseaseIndicator'])['BeneID'].count())

	BeneID
RenalDiseaseIndicator
0	118978
Y	19578

RKD_IP_R = pd.DataFrame(train_bene_df.groupby(['RenalDiseaseIndicator'])['IPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['RenalDiseaseIndicator'])['BeneID'].count())
RKD_IP_R.columns = ['AVG IP Reimbursement Amt']
RKD_IP_R

	AVG IP Reimbursement Amt
RenalDiseaseIndicator
0	3051.430264
Y	7360.808050

RKD_OP_R = pd.DataFrame(train_bene_df.groupby(['RenalDiseaseIndicator'])['OPAnnualReimbursementAmt'].sum() / train_bene_df.groupby(['RenalDiseaseIndicator'])['BeneID'].count())
RKD_OP_R.columns = ['AVG OP Reimbursement Amt']
RKD_OP_R

	AVG OP Reimbursement Amt
RenalDiseaseIndicator
0	1047.351611
Y	2822.774543

RKD_IP_D = pd.DataFrame(train_bene_df.groupby(['RenalDiseaseIndicator'])['IPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['RenalDiseaseIndicator'])['BeneID'].count())
RKD_IP_D.columns = ['AVG IP Co-payment Amt']
RKD_IP_D

	AVG IP Co-payment Amt
RenalDiseaseIndicator
0	342.384239
Y	749.057616

RKD_OP_D = pd.DataFrame(train_bene_df.groupby(['RenalDiseaseIndicator'])['OPAnnualDeductibleAmt'].sum() / train_bene_df.groupby(['RenalDiseaseIndicator'])['BeneID'].count())
RKD_OP_D.columns = ['AVG OP Co-payment Amt']
RKD_OP_D

	AVG OP Co-payment Amt
RenalDiseaseIndicator
0	311.659139
Y	779.167944

RKD_all_amts = pd.concat([RKD_IP_R, RKD_OP_R, RKD_IP_D, RKD_OP_D], axis=1)
RKD_all_amts

	AVG IP Reimbursement Amt	AVG OP Reimbursement Amt	AVG IP Co-payment Amt	AVG OP Co-payment Amt
RenalDiseaseIndicator
0	3051.430264	1047.351611	342.384239	311.659139
Y	7360.808050	2822.774543	749.057616	779.167944

# Here, I'm displaying the Total Annual Sum of IP Co-payment for 'RenalDiseaseIndicator'
with plt.style.context('seaborn-poster'):
  fig = RKD_all_amts.plot(kind='bar', colormap='rainbow')
  # Using the "patches" function we will get the location of the rectangle bars from the graph.
  ## Then by using those location(width & height) values we will add the annotations
  for p in fig.patches:
    width = p.get_width()
    height = p.get_height()
    x, y = p.get_xy()
    fig.annotate(f'{round(height,0)}', (x + width/2, y + height*1.015), ha='center', fontsize=12, rotation=0)
  # Providing the labels and title to the graph
  plt.xlabel("\nHaving Chronic Renal Kidney Disease?", fontdict=label_font_dict)
  plt.xticks(ticks=[0,1], labels=['NO', 'YES'], fontsize=13, rotation=30)
  plt.ylabel("Total Annual Sum \n", fontdict=label_font_dict)
  plt.grid(which='major', linestyle="-.", color='lightgrey')
  plt.minorticks_on()
  plt.title("Total Annual Sum of various amounts : 'RenalDiseaseIndicator'\n", fontdict=title_font_dict)

# Y means +ve with Renal Kidney Disease
# 0 means -ve with Renal Kidney Disease

• OBSERVATION
  • The above graph is telling us below points:
    • Payer pays huge chunk of expenses specially when a beneficiary gets admitted with or without Renal Kidney Disease. To be more precise, in case of +ve the difference is more than 50%.
    • For other comparisons the difference is not very high.

Q21. Lets check the percentiles of the pre-disease indicators for the Annual IP expenditures for such patients.

def cal_display_percentiles(x_col, y_col, title_lbl, x_filter_code):
    """
    Description : This function is created for calculating and generating the percentiles for pre-disease indicators.
    
    Input: It accepts below parameters:
        1. x_col : Disease indicator feature name.
        2. y_col : Feature like re-imbursement or deductible amount whose percentiles you want to generate.
        3. title_lbl : Label to be provided in the title of the plot.
        4. x_filter_code : Category code for which you want to generate the percentiles.
        
    Output: It returns the dataframe having percentiles and their respective values for the specific disease indicator feature.
    And, it displays the pointplot graph of the same.
    """
    percentiles = []
    percentiles_vals = []

    # Calculating & storing the various percentiles and their respective values
    for val in [0.1,0.2,0.25,0.3,0.4,0.5,0.6,0.7,0.75,0.8,0.9,0.91,0.92,0.93,0.94,0.95,0.96,0.97,0.98,0.99,0.999,0.9999,0.99999,0.999999,1.0]:
        percentile = round(float(val*100),6)
        percentiles.append(percentile)

        percentile_val = round(train_bene_df[train_bene_df[x_col] == x_filter_code][y_col].quantile(val),1)
        percentiles_vals.append(percentile_val)

    # Creating the temp dataframe for displaying the results
    tmp_percentiles = pd.DataFrame([percentiles, percentiles_vals]).T
    tmp_percentiles.columns = ['Percentiles', 'Values']

    # Here, I'm displaying the Percentiles values for all disease code features
    with plt.style.context('seaborn-poster'):
        plt.figure(figsize=(15,7))
        sns.pointplot(data=tmp_percentiles, x='Percentiles', y='Values', markers="o", palette='spring')
        sns.pointplot(data=tmp_percentiles, x='Percentiles', y='Values', markers="", color='grey', linestyles="solid")
        # Providing the labels and title to the graph
        plt.xlabel("\nPercentiles", fontdict=label_font_dict)
        plt.xticks(rotation=90, size=12)
        plt.ylabel("Total Annual `{}` Sum \n".format(y_col), fontdict=label_font_dict)
        plt.grid(which='major', linestyle="-.", color='lightpink')
        plt.minorticks_on()
        plt.title("Percentile values of `{}` :: `{}`\n".format(y_col,title_lbl), fontdict=title_font_dict)
        
    return tmp_percentiles

• (RenalDiseaseIndicator == YES) and (IPAnnualReimbursementAmt)

RKD_YES_IP_R_percentiles = cal_display_percentiles(x_col='RenalDiseaseIndicator', 
                                                   y_col='IPAnnualReimbursementAmt',
                                                   title_lbl="Renal Kidney Disease = YES",
                                                   x_filter_code='Y')

• OBSERVATION
  • The above graph shows us that some of the reimbursements paid by the PAYER are very high as compared to the rest of the records.
    • This can be a potential sign of fraudulent cases because generally the criminals file some forge cases with exponentially high amounts.

• (RenalDiseaseIndicator == NO) and (IPAnnualReimbursementAmt)

RKD_NO_IP_R_percentiles = cal_display_percentiles(x_col='RenalDiseaseIndicator', 
                                                   y_col='IPAnnualReimbursementAmt',
                                                   title_lbl="Renal Kidney Disease = NO",
                                                   x_filter_code='0')

• OBSERVATION
  • The above graph shows us that some of the reimbursements paid by the PAYER are very high as compared to the rest of the records.
    • This can be a potential sign of fraudulent cases because generally the criminals file some forge cases with exponentially high amounts.

• (RenalDiseaseIndicator == YES) and (OPAnnualReimbursementAmt)

RKD_YES_OP_R_percentiles = cal_display_percentiles(x_col='RenalDiseaseIndicator', 
                                                   y_col='OPAnnualReimbursementAmt',
                                                   title_lbl="Renal Kidney Disease = YES",
                                                   x_filter_code='Y')

• OBSERVATION
  • The above graph shows us that some of the reimbursements paid by the PAYER are very high as compared to the rest of the records.
    • This can be a potential sign of fraudulent cases because generally the criminals file some forge cases with exponentially high amounts.

• (RenalDiseaseIndicator == NO) and (OPAnnualReimbursementAmt)

RKD_NO_OP_R_percentiles = cal_display_percentiles(x_col='RenalDiseaseIndicator', 
                                                   y_col='OPAnnualReimbursementAmt',
                                                   title_lbl="Renal Kidney Disease = NO",
                                                   x_filter_code='0')

• OBSERVATION
  • The above graph shows us that some of the reimbursements paid by the PAYER are very high as compared to the rest of the records.
    • This can be a potential sign of fraudulent cases because generally the criminals file some forge cases with exponentially high amounts.

• (ChronicCond_stroke == YES) and (IPAnnualReimbursementAmt)

CC_ST_YES_IP_R_percentiles = cal_display_percentiles(x_col='ChronicCond_stroke', 
                                                     y_col='IPAnnualReimbursementAmt',
                                                     title_lbl="ChronicCond_stroke = YES",
                                                     x_filter_code=1)

• OBSERVATION
  • The above graph shows us that some of the reimbursements paid by the PAYER are very high as compared to the rest of the records.
    • This can be a potential sign of fraudulent cases because generally the criminals file some forge cases with exponentially high amounts.

• (ChronicCond_stroke == NO) and (IPAnnualReimbursementAmt)

CC_ST_NO_IP_R_percentiles = cal_display_percentiles(x_col='ChronicCond_stroke', 
                                                     y_col='IPAnnualReimbursementAmt',
                                                     title_lbl="ChronicCond_stroke = NO",
                                                     x_filter_code=2)

• OBSERVATION
  • The above graph shows us that some of the reimbursements paid by the PAYER are very high as compared to the rest of the records.
    • This can be a potential sign of fraudulent cases because generally the criminals file some forge cases with exponentially high amounts.

• (ChronicCond_stroke == YES) and (OPAnnualReimbursementAmt)

CC_ST_YES_OP_R_percentiles = cal_display_percentiles(x_col='ChronicCond_stroke', 
                                                     y_col='OPAnnualReimbursementAmt',
                                                     title_lbl="ChronicCond_stroke = YES",
                                                     x_filter_code=1)

• OBSERVATION
  • The above graph shows us that some of the reimbursements paid by the PAYER are very high as compared to the rest of the records.
    • This can be a potential sign of fraudulent cases because generally the criminals file some forge cases with exponentially high amounts.

• (ChronicCond_stroke == NO) and (OPAnnualReimbursementAmt)

CC_ST_NO_OP_R_percentiles = cal_display_percentiles(x_col='ChronicCond_stroke', 
                                                     y_col='OPAnnualReimbursementAmt',
                                                     title_lbl="ChronicCond_stroke = NO",
                                                     x_filter_code=2)

• OBSERVATION
  • The above graph shows us that some of the reimbursements paid by the PAYER are very high as compared to the rest of the records.
    • This can be a potential sign of fraudulent cases because generally the criminals file some forge cases with exponentially high amounts.

Q22. Lets just visualize the spread of pre-disease indicators for the Annual IP and OP expenditures across males and females.

def plot_strip_plots(x_col, hue_col, y_col, lgd_title):
    """
    Description : This function is created for plotting the spread of data points of pre-disease indicators for the Annual IP and OP expenditures
    across males and females.
    
    Input: It accepts below parameters:
        1. x_col : Gender feature.
        2. hue_Col : Pre-Disease indicator
        3. y_col : Feature like re-imbursement or deductible amount whose percentiles you want to generate.
        4. lgd_title : Category code for which you want to generate the data spread.
        
    Output: It displays the stipplot graph of the same.
    """
    with plt.style.context('seaborn-poster'):
        plt.figure(figsize=(10,7))
        sns.stripplot(data=train_bene_df, x=x_col, y=y_col, hue=hue_col, palette='plasma')
        # Providing the labels and title to the graph
        plt.xlabel("\n{}".format(x_col), fontdict=label_font_dict)
        plt.xticks(rotation=90, size=12)
        plt.ylabel("{}\n".format(y_col), fontdict=label_font_dict)
        plt.grid(which='major', linestyle="-.", color='lightpink')
        plt.minorticks_on()
        plt.title("Spread of payment paid by payer\n", fontdict=title_font_dict)
        plt.legend(loc='upper center',title=lgd_title)

• (RenalDiseaseIndicator) , (IPAnnualReimbursementAmt) and (GENDER)

plot_strip_plots(x_col='Gender', hue_col="RenalDiseaseIndicator", y_col='IPAnnualReimbursementAmt', lgd_title="Renal Kidney Disease")

• OBSERVATION
  • The above graph shows us the complete overlapping of data points with some potential outliers(may be fraud).
    • Here, another thing that I found is that few of the points lying in the negative range(this is quite strange may be error).

train_bene_df['OPAnnualReimbursementAmt'].min(), train_bene_df['IPAnnualReimbursementAmt'].min()

(-70, -8000)

train_bene_df['OPAnnualDeductibleAmt'].min(), train_bene_df['IPAnnualDeductibleAmt'].min()

(0, 0)

• (RenalDiseaseIndicator) , (OPAnnualReimbursementAmt) and (GENDER)

plot_strip_plots(x_col='Gender', hue_col="RenalDiseaseIndicator", y_col='OPAnnualReimbursementAmt', lgd_title="Renal Kidney Disease")

• OBSERVATION
  • The above graph shows us the complete overlapping of data points with some potential outliers(may be fraud).
    • Here, another thing taht I found is that few of the points lying in the negative range(this is quite strange may be error).

• (ChronicCond_rheumatoidarthritis) , (IPAnnualReimbursementAmt) and (GENDER)

plot_strip_plots(x_col='Gender', hue_col="ChronicCond_rheumatoidarthritis", y_col='IPAnnualReimbursementAmt', lgd_title="Rheumatoidarthritis")

• OBSERVATION
  • The above graph shows us the complete overlapping of data points with some potential outliers(may be fraud).
    • Here, another thing taht I found is that few of the points lying in the negative range(this is quite strange may be error).

• (ChronicCond_rheumatoidarthritis) , (OPAnnualReimbursementAmt) and (GENDER)

plot_strip_plots(x_col='Gender', hue_col="ChronicCond_rheumatoidarthritis", y_col='OPAnnualReimbursementAmt', lgd_title="Rheumatoidarthritis")

• OBSERVATION
  • The above graph shows us the complete overlapping of data points with some potential outliers(may be fraud).
    • Here, another thing taht I found is that few of the points lying in the negative range(this is quite strange may be error).

• (ChronicCond_IschemicHeart) , (IPAnnualReimbursementAmt) and (GENDER)

plot_strip_plots(x_col='Gender', hue_col="ChronicCond_IschemicHeart", y_col='IPAnnualReimbursementAmt', lgd_title="Ischemic Heart")

• OBSERVATION
  • The above graph shows us the complete overlapping of data points with some potential outliers specifically for patients with Chronic Heart Desiase(may be fraud).
    • Here, another thing taht I found is that few of the points lying in the negative range(this is quite strange may be error).

• (ChronicCond_IschemicHeart) , (OPAnnualReimbursementAmt) and (GENDER)

plot_strip_plots(x_col='Gender', hue_col="ChronicCond_IschemicHeart", y_col='OPAnnualReimbursementAmt', lgd_title="Ischemic Heart")

• OBSERVATION
  • The above graph shows us the complete overlapping of data points with some potential outliers specifically for patients with Chronic Heart Desiase(may be fraud).
    • Here, another thing taht I found is that few of the points lying in the negative range(this is quite strange may be error).

Q23. Lets visualize the spread of Annual IP and OP expenditures through out the AGE and its assciated features for males and females.

with plt.style.context('seaborn-poster'):
    plt.figure(figsize=(12,12))
    sns.scatterplot(data=train_bene_df, x='AGE', y='IPAnnualReimbursementAmt', hue='Gender', palette='cubehelix')
    plt.grid(which='major', linestyle="-.", color='lightpink')
    plt.minorticks_on()
    plt.title("Spread of payment paid by payer for entire age\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us the complete overlapping of data points.

with plt.style.context('seaborn-poster'):
    plt.figure(figsize=(12,12))
    sns.scatterplot(data=train_bene_df, x='AGE', y='OPAnnualReimbursementAmt', hue='Gender', palette='cubehelix')
    plt.grid(which='major', linestyle="-.", color='lightpink')
    plt.minorticks_on()
    plt.title("Spread of payment paid by payer for entire age\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us the complete overlapping of data points.

• (AGE GROUPS) , (IPAnnualReimbursementAmt) and (GENDER)

with plt.style.context('seaborn-poster'):
    plt.figure(figsize=(12,12))
    sns.boxenplot(data=train_bene_df, x='AGE_groups', y='IPAnnualReimbursementAmt', hue='Gender', palette='cubehelix')
    plt.minorticks_on()
    plt.title("Spread of payment paid by payer across age groups\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us there is no such difference in the amounts across different AGE Groups.

• (AGE GROUPS) , (OPAnnualReimbursementAmt) and (GENDER)

with plt.style.context('seaborn-poster'):
    plt.figure(figsize=(12,12))
    sns.boxenplot(data=train_bene_df, x='AGE_groups', y='OPAnnualReimbursementAmt', hue='Gender', palette='cubehelix')
    plt.minorticks_on()
    plt.title("Spread of payment paid by payer across age groups\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us there is no such difference in the amounts across different AGE Groups.

• (DOB MONTH) , (IPAnnualReimbursementAmt) and (GENDER)

with plt.style.context('seaborn-poster'):
    plt.figure(figsize=(16,12))
    sns.boxenplot(data=train_bene_df, x='Patient_Age_Month', y='IPAnnualReimbursementAmt', hue='Gender', palette='cubehelix')
    plt.minorticks_on()
    plt.title("Spread of payment paid by payer across DOB Months\n", fontdict=title_font_dict)
    plt.legend(loc='upper right', title="Gender")

• OBSERVATION
  • The above graph shows us there is no such difference in the amounts across different AGE Groups.

• (DOB MONTH) , (OPAnnualReimbursementAmt) and (GENDER)

with plt.style.context('seaborn-poster'):
    plt.figure(figsize=(16,12))
    sns.boxenplot(data=train_bene_df, x='Patient_Age_Month', y='OPAnnualReimbursementAmt', hue='Gender', palette='cubehelix')
    plt.minorticks_on()
    plt.title("Spread of payment paid by payer across DOB Months\n", fontdict=title_font_dict)
    plt.legend(loc='upper right', title="Gender")

• OBSERVATION
  • The above graph shows us there is no such difference in the amounts across different AGE Groups.

• (DOB YEARS) , (IPAnnualReimbursementAmt) and (GENDER)

with plt.style.context('seaborn'):
    plt.figure(figsize=(16,12))
    sns.stripplot(data=train_bene_df, x='Patient_Age_Year', y='IPAnnualReimbursementAmt', hue='Gender', palette='cubehelix')
    plt.xticks(rotation=90, fontsize=11)
    plt.grid(which='major', linestyle="-.", color='lightpink')
    plt.minorticks_on()
    plt.title("Spread of payment paid by payer across DOB Years\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us there is no such difference in the amounts across different AGE Groups.

• (DOB YEARS) , (OPAnnualReimbursementAmt) and (GENDER)

with plt.style.context('seaborn'):
    plt.figure(figsize=(16,12))
    sns.stripplot(data=train_bene_df, x='Patient_Age_Year', y='OPAnnualReimbursementAmt', hue='Gender', palette='cubehelix')
    plt.xticks(rotation=90, fontsize=11)
    plt.grid(which='major', linestyle="-.", color='lightpink')
    plt.minorticks_on()
    plt.title("Spread of payment paid by payer across DOB Years\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us there is no such difference in the amounts across different AGE Groups.

• (LIFE Status) , (IPAnnualReimbursementAmt) and (GENDER)

with plt.style.context('seaborn-poster'):
    plt.figure(figsize=(12,8))
    sns.boxenplot(data=train_bene_df, x='Dead_or_Alive', y='IPAnnualReimbursementAmt', hue='Gender', palette='autumn')
    plt.minorticks_on()
    plt.title("Spread of payment paid by payer based on life status\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us there is no such difference in the amounts across different AGE Groups.

• (LIFE STATUS) , (OPAnnualReimbursementAmt) and (GENDER)

with plt.style.context('seaborn-poster'):
    plt.figure(figsize=(12,8))
    sns.boxenplot(data=train_bene_df, x='Dead_or_Alive', y='OPAnnualReimbursementAmt', hue='Gender', palette='autumn')
    plt.minorticks_on()
    plt.title("Spread of payment paid by payer based on life status\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us there is no such difference in the amounts across different AGE Groups.

• (HUMAN RACE) , (IPAnnualReimbursementAmt) and (GENDER)

with plt.style.context('seaborn-poster'):
    plt.figure(figsize=(12,8))
    sns.boxenplot(data=train_bene_df, x='Race', y='IPAnnualReimbursementAmt', hue='Gender', palette='twilight')
    plt.minorticks_on()
    plt.title("Spread of payment paid by payer based on life status\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us there is no such difference in the amounts across different AGE Groups.

• (HUMAN RACE) , (OPAnnualReimbursementAmt) and (GENDER)

with plt.style.context('seaborn-poster'):
    plt.figure(figsize=(12,8))
    sns.boxenplot(data=train_bene_df, x='Race', y='OPAnnualReimbursementAmt', hue='Gender', palette='twilight')
    plt.minorticks_on()
    plt.title("Spread of payment paid by payer based on life status\n", fontdict=title_font_dict)

• OBSERVATION
  • The above graph shows us there is no such difference in the amounts across different AGE Groups.

SUMMARY

1. For the below mentioned features, based on the above initial analysis it looks like these might not be able to provide much information or differentiation but still I would like to check them after adding CLAIMS data.
   • DOB YEAR
   • DOB MONTH
   • AGE GROUPS
   • LIFE STATUS
   • HUMAN RACE
   • STATE

1. For the below mentioned features majority of the values are same which most probably won't be of any use thus removing these from BENE dataset.
   • NoOfMonths_PartACov
   • NoOfMonths_PartBCov

1. The Pre-disease indicators looks like important features based on the initial analysis thus it would interesting to see how much they are useful after adding CLAIMS dataset.

1. Date of Death is also removed from the dataset, as we have already calculated bene age, life status and others out of it.

train_bene_df.drop(["NoOfMonths_PartACov", "NoOfMonths_PartBCov"], axis=1, inplace=True)

train_bene_df.shape

(138556, 27)

train_bene_df.head()

	BeneID	DOB	Gender	Race	RenalDiseaseIndicator	State	County	ChronicCond_Alzheimer	ChronicCond_Heartfailure	ChronicCond_KidneyDisease	ChronicCond_Cancer	ChronicCond_ObstrPulmonary	ChronicCond_Depression	ChronicCond_Diabetes	ChronicCond_IschemicHeart	ChronicCond_Osteoporasis	ChronicCond_rheumatoidarthritis	ChronicCond_stroke	IPAnnualReimbursementAmt	IPAnnualDeductibleAmt	OPAnnualReimbursementAmt	OPAnnualDeductibleAmt	Patient_Age_Year	Patient_Age_Month	Dead_or_Alive	AGE	AGE_groups
0	BENE11001	1943-01-01	1	1	0	39	230	1	2	1	2	2	1	1	1	2	1	1	36000	3204	60	70	1943	1	0	67.0	Old
1	BENE11002	1936-09-01	0	1	0	39	280	2	2	2	2	2	2	2	2	2	2	2	0	0	30	50	1936	9	0	73.3	Old
2	BENE11003	1936-08-01	1	1	0	52	590	1	2	2	2	2	2	2	1	2	2	2	0	0	90	40	1936	8	0	73.4	Old
3	BENE11004	1922-07-01	1	1	0	39	270	1	1	2	2	2	2	1	1	1	1	2	0	0	1810	760	1922	7	0	87.5	Very Old
4	BENE11005	1935-09-01	1	1	0	24	680	2	2	2	2	1	2	1	2	2	2	2	0	0	1790	1200	1935	9	0	74.3	Old

train_bene_df.to_csv("train_bene_1.csv")